Gene family proliferation has provided raw material for functional innovation in higher eukaryotes. It is important to understand the functional divergence among homologous gene clusters in a gene family. Predicting functional divergence from sequence information is one of the challenge tasks in functional divergence. We developed a statistic model to estimate the level of functional divergence and further predict critical amino acid residues. The studies in vertebrate gene families show that functional divergence is a general pattern, and the three components of our new method PHYBA (coefficient of functional divergence, site-specific profile, and functional branch length) can make statistically sound and biologically meaningful predictions. In particular, the case study in Caspase gene family has shown the structural base of functional divergence.;Functional divergence in gene families is highly correlated with the gene duplication, which provides opportunity for generating novelty of protein function. We investigate 49 vertebrate gene families that were generated in the early stage of vertebrates, each of which consists of three or four member genes. The age distribution of these gene families suggest that two round of large-scale gene duplications is still a valid working hypothesis: the means of the first and the second gene duplication times are 594 mya and 488 mya. Moreover, these two duplication events seem to be independent.